\
OpenAI announced the launch of Codex, a cloud-based AI coding tool designed to streamline and automate various software development tasks. Codex can handle everything from generating new features and fixing bugs to answering questions about codebases and suggesting pull requests. Unlike earlier tools, Codex functions within its own cloud sandbox environment that is preloaded with the user’s code repository, enabling it to execute tasks in a contextually aware and isolated setup. While the tool is currently available under research preview, all ChatGPT Pro, Enterprise, and Team users have been granted early access. OpenAI noted that users will have generous access at no additional cost for the coming weeks before a rate-limited pricing model is introduced. ChatGPT Plus and Edu users are expected to receive access at a later date.
Codex is powered by codex-1, a variant of OpenAI’s o3 reasoning model that has been fine-tuned specifically for software engineering. Trained on real-world programming tasks, codex-1 generates code that closely mirrors human standards and pull request preferences. The model has been further refined through reinforcement learning, allowing it to iteratively run tests until successful outcomes are achieved. OpenAI reported that codex-1 outperforms o3 on internal and human-validated coding benchmarks.
In practice, Codex can read, edit, and run code, executing commands such as linters, test harnesses, and type checkers. Depending on task complexity, completion can take between one to thirty minutes. Codex also supports an AGENTS.md file, similar to README.md, which users can include in their repositories to help guide the AI through their codebase, including test instructions and coding standards. A standout feature is Codex’s ability to show its reasoning in real-time, offering terminal logs and test outputs as citations so developers can trace each step.
To use Codex, users input prompts and choose between the ‘code’ or ‘ask’ options, depending on whether they want code generation or contextual answers. Early testers have used Codex for tasks like debugging, test writing, code refactoring, and documentation. OpenAI recommends assigning well-scoped tasks to multiple agents simultaneously and experimenting with varied prompts to make the most of Codex’s capabilities.
The company also differentiated Codex from Codex CLI, a command-line tool released earlier in April. Codex CLI is an open-source terminal-based agent that integrates OpenAI’s models with a user’s local development environment. Powered by the o4-mini model (customizable via API), Codex CLI currently supports macOS and Linux, with Windows support still under development. Recent updates include the introduction of a smaller codex-1 model for Codex CLI (branded as codex-mini-latest) and a simplified login process using ChatGPT credentials. Eligible Plus and Pro users signing in through ChatGPT can also redeem $5 and $50 in free API credits, respectively, valid for the next 30 days.
OpenAI’s release of Codex arrives at a pivotal moment as AI continues to reshape the software engineering industry. With tech leaders like Microsoft CEO Satya Nadella stating that over 30% of their code is now AI-generated, tools like Codex underscore both the potential and challenges of integrating AI in development workflows—particularly the importance of human oversight in reviewing AI-generated code.
